ffmpeg-v4l2-request-fourier: PKGREL 3 → 5 (force rebuild past orphan -4 .deb)

PR #76 (H.264 IDCT 4×4 daedalus-fourier substitution) was merged but
the resulting .deb was not actually built: an orphan
ffmpeg-v4l2-request-fourier_8.1+rfourier+gb57fbbe-4_arm64.deb (dated
2026-05-19, no matching source commit in main) sat in the apt pool.
.gitea/scripts/check-already-published.sh's debian branch compares
`dpkg --compare-versions $pool_ver ge $source_full` — pool -4
≥ source -3, so CI's skip-check emitted skip=1 and short-circuited
the build.  The ffmpeg-v4l2-request-debian Action reported success
without actually publishing.

Bump source PKGREL past -4 so the next CI run sees source >= pool
and proceeds to build + publish.

No source code change beyond PKGREL + changelog.  Arch side
unaffected (its skip-check is exact-URL-match, not pool-head-ge).

.gitea/workflows/build.yml

@@ -930,12 +930,13 @@ jobs:
           # map 1:1 to the previous Arch list; libav*-dev intentionally
           # absent (we are FFmpeg itself, providing those libs).
           retry apt-get install -y --no-install-recommends \
-              build-essential git pkg-config nasm yasm \
+              build-essential cmake ninja-build git pkg-config nasm yasm \
               linux-libc-dev libgl1-mesa-dev libasound2-dev libbz2-dev \
               libfontconfig-dev libfribidi-dev libgmp-dev libgnutls28-dev \
               libmp3lame-dev libass-dev libdav1d-dev libdrm-dev \
               libfreetype-dev libpulse-dev libva-dev libvorbis-dev libvpx-dev \
               libwebp-dev libx264-dev libx265-dev libxml2-dev libopus-dev \
+              libvulkan-dev glslang-tools \
               v4l-utils liblzma-dev zlib1g-dev \
               curl ca-certificates openssh-client rsync dpkg-dev
 
@@ -1159,16 +1160,30 @@ jobs:
           retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
           export DEBIAN_FRONTEND=noninteractive
           retry apt-get update -qq
-          # libav*-dev provide the headers daedalus daemon dlopens at
-          # runtime — Debian's stock packages match the trixie ABI the
-          # daemon will encounter on Pi 5 hosts (both ship libavcodec
-          # 61.x).  The fourier ffmpeg fork isn't needed here; the
-          # daemon never link-binds against libav (Option γ — dlopen
-          # at runtime), so any header set with the right struct
-          # definitions works.
+          # FFmpeg headers + sonames the daemon dlopens.  As of
+          # daedalus-v4l2 PR #16 (commit 514da29), the daemon targets
+          # the Kwiboo fork's libavcodec.so.62 / libavformat.so.62 /
+          # libavutil.so.60 at /opt/fourier — so the build needs
+          # /opt/fourier/include and /opt/fourier/lib/pkgconfig.
+          # ffmpeg-v4l2-request-fourier provides both (plus the
+          # runtime libs the .deb will dlopen on the target host;
+          # we install it as a build-dep here and the dpkg-shlibdeps
+          # step pulls it into the daemon .deb's Depends automatically).
+          # Debian-stock libav*-dev removed — would conflict on
+          # /usr/include/libavcodec/avcodec.h vs /opt/fourier's copy.
+          #
+          # libvulkan-dev + glslang-tools: needed by the in-build
+          # daedalus-fourier fetch (build-deb.sh fetches the sibling
+          # library, cmake-builds it into a temp prefix, then the
+          # daedalus daemon static-links against it via pkg-config).
+          # Without these, daedalus-fourier's find_package(Vulkan)
+          # and glslangValidator find_program both fail at configure
+          # time.  See marfrit/daedalus-fourier PR #1 +
+          # reauktion/daedalus-v4l2 PR #13.
           retry apt-get install -y --no-install-recommends \
               build-essential cmake ninja-build pkg-config git \
-              libavcodec-dev libavformat-dev libavutil-dev libdrm-dev \
+              ffmpeg-v4l2-request-fourier libdrm-dev \
+              libvulkan-dev glslang-tools \
               linux-libc-dev \
               curl ca-certificates openssh-client rsync dpkg-dev
 

arch/daedalus-v4l2-dkms/PKGBUILD

@@ -18,12 +18,15 @@ _module=daedalus_v4l2
 
 # Same pin as arch/daedalus-v4l2 — keep kernel module + daemon
 # bit-versioned together so the chardev wire protocol stays in sync.
-# PROTO_VERSION 0 → 1 at this pin (H.264 B-frame reorder fix); must
-# install both packages atomically.
-_commit=79256dc7ef41f83873ca9c23db20f5888858e65d
+# 5d8b436 reverts PRs #7 + #8 (parking design that broke libva's
+# 1:1 contract — see daedalus-v4l2#9 + #10).  Tree is
+# content-equivalent to f0d4186 plus PR #4 (cosmetic menu ctrls).
+# PROTO_VERSION drops 1 → 0; lock-step install with
+# daedalus-v4l2 0.1.0.r33.5d8b436 REQUIRED.
+_commit=5d8b4369e58ab947d1c56b1f718293c57c6065b5
 
-pkgver=0.1.0.r28.79256dc
-pkgrel=1  # reset for new upstream pin (79256dc — H.264 B-frame reorder fix)
+pkgver=0.1.0.r33.5d8b436
+pkgrel=1  # reset for new upstream pin (5d8b436 — revert parking design)
 pkgdesc="V4L2 stateless decoder shim kernel module (DKMS) — Pi 5 / CM5"
 arch=('any')
 url="https://git.reauktion.de/reauktion/daedalus-v4l2"

arch/daedalus-v4l2/PKGBUILD

@@ -16,18 +16,19 @@
 pkgname=daedalus-v4l2
 _upstreampkg=daedalus-v4l2
 
-# Pin the daedalus-v4l2 tip.  79256dc = "kernel + daemon: H.264 B-frame
-# display reorder fix (closes #6)" — adds the wire-protocol src_pts /
-# output_src_pts / RESP_FRAME flags split that lets H.264 streams with
-# B-frames preserve display order through libva → kernel → daemon.
-# PROTO_VERSION bumps 0 → 1; lock-step userspace + kernel rebuild
-# REQUIRED (daedalus-v4l2-dkms PKGBUILD pinned to the same commit).
-_commit=79256dc7ef41f83873ca9c23db20f5888858e65d
+# 6e6dfa1 = picks up daedalus-v4l2 PR #16 — daemon now dlopens
+# the Kwiboo fourier fork's libavcodec.so.62 / libavformat.so.62 /
+# libavutil.so.60 at /opt/fourier instead of Debian-stock soname
+# 61/61/59.  First step on the daedalus-fourier substitution arc
+# (daedalus-v4l2#11).  Daemon still needs daedalus-fourier at
+# build time (Arch packaging for that is a follow-up; Debian side
+# fetches inline via build-deb.sh).
+_commit=6e6dfa144da7bc7fa8be50c8da91d7d1c6132a2c
 
 # 0.1.0 (pre-1.0) + commit count + short sha.  Bump the .Y on each
 # Phase 8.x close.  pkgver() recomputes at build time.
-pkgver=0.1.0.r28.79256dc
-pkgrel=1  # reset for new upstream pin (79256dc — H.264 B-frame reorder fix)
+pkgver=0.1.0.r41.6e6dfa1
+pkgrel=1  # reset for new upstream pin (6e6dfa1 — soname 62 via /opt/fourier)
 pkgdesc="Userspace daemon for the daedalus-v4l2 V4L2 stateless decoder shim (VP9/AV1/H.264 on Pi 5 / CM5)"
 arch=('aarch64')
 url="https://git.reauktion.de/reauktion/daedalus-v4l2"
@@ -35,7 +36,7 @@ license=('BSD-2-Clause' 'GPL-2.0-or-later')
 # Daemon dlopens libavformat.so.61 / libavcodec.so.61 / libavutil.so.59
 # at runtime (Option γ — see daemon/src/ffmpeg_loader.h).  ffmpeg
 # provides those; we don't link them.
-depends=('ffmpeg' 'libdrm')
+depends=('ffmpeg-v4l2-request-fourier' 'libdrm')
 # Headers from libav*-dev needed at compile time for type-safe function
 # pointer signatures; pkg-config locates them.
 makedepends=('cmake' 'ninja' 'pkgconf' 'git' 'ffmpeg')

arch/ffmpeg-v4l2-request-fourier/0003-h264-idct4-daedalus-fourier.patch

@@ -0,0 +1,137 @@
+From f760c0541586f43334c02611fcb4c212c08ad576 Mon Sep 17 00:00:00 2001
+From: Markus Fritsche <mfritsche@reauktion.de>
+Date: Thu, 21 May 2026 21:40:22 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 4x4 IDCT through
+ daedalus-fourier
+MIME-Version: 1.0
+Content-Type: text/plain; charset=UTF-8
+Content-Transfer-Encoding: 8bit
+
+H264DSPContext.idct_add (called per 4x4 block from the intra-4x4
+decode path in h264_mb.c) now dispatches through
+daedalus_recipe_dispatch_h264_idct4 instead of ff_h264_idct_add_neon.
+
+The recipe layer picks the substrate; for cycle 6 (H.264 IDCT 4x4)
+the recipe is CPU NEON, so this is effectively a NEON-to-NEON
+substitution with one extra dispatch call and recipe-table lookup.
+Provides the first end-to-end exercise of the daedalus-fourier
+kernel pack inside the libavcodec.so decode hot path; follow-up
+patches wire IDCT 8x8, luma-v deblock, and qpel mc20.
+
+The library context is process-global, lazily initialised under
+pthread_once on first call.  We pick the no-QPU constructor because
+libavcodec.so is loaded into arbitrary host processes
+(firefox-fourier, mpv-fourier, daedalus_v4l2_daemon, ...) and we
+cannot assume the host has a usable Vulkan instance.  Higher cycles
+(deblock luma-v, MC) that benefit from the QPU will provision their
+own recipe-selected context once that path is wired.
+
+Bulk paths (idct_add16, idct_add16intra, idct_add8 — used for
+non-intra4x4 macroblocks) remain on the stock NEON .S implementations
+and will be batched through daedalus_recipe_dispatch_h264_idct4 with
+n_blocks>1 in a follow-up.
+
+Bit-exact against ff_h264_idct_add_neon (daedalus-fourier cycle 6
+green; see marfrit/daedalus-fourier/CYCLE_LOGS.md).
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2.
+---
+ libavcodec/aarch64/Makefile               |  3 +-
+ libavcodec/aarch64/h264_idct_daedalus.c   | 49 +++++++++++++++++++++++
+ libavcodec/aarch64/h264dsp_init_aarch64.c |  3 +-
+ 3 files changed, 53 insertions(+), 2 deletions(-)
+ create mode 100644 libavcodec/aarch64/h264_idct_daedalus.c
+
+diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
+index 41ab025..7b95fb1 100644
+--- a/libavcodec/aarch64/Makefile
++++ b/libavcodec/aarch64/Makefile
+@@ -3,7 +3,8 @@ OBJS-$(CONFIG_AC3DSP)                   += aarch64/ac3dsp_init_aarch64.o
+ OBJS-$(CONFIG_FDCTDSP)                  += aarch64/fdctdsp_init_aarch64.o
+ OBJS-$(CONFIG_FMTCONVERT)               += aarch64/fmtconvert_init.o
+ OBJS-$(CONFIG_H264CHROMA)               += aarch64/h264chroma_init_aarch64.o
+-OBJS-$(CONFIG_H264DSP)                  += aarch64/h264dsp_init_aarch64.o
++OBJS-$(CONFIG_H264DSP)                  += aarch64/h264dsp_init_aarch64.o \
++                                           aarch64/h264_idct_daedalus.o
+ OBJS-$(CONFIG_HUFFYUVDSP)               += aarch64/huffyuvdsp_init_aarch64.o
+ OBJS-$(CONFIG_H264PRED)                 += aarch64/h264pred_init.o
+ OBJS-$(CONFIG_H264QPEL)                 += aarch64/h264qpel_init_aarch64.o
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+new file mode 100644
+index 0000000..538d223
+--- /dev/null
++++ b/libavcodec/aarch64/h264_idct_daedalus.c
+@@ -0,0 +1,49 @@
++/*
++ * H.264 4x4 IDCT + add — daedalus-fourier substitution shim.
++ *
++ * Routes H264DSPContext.idct_add through
++ * daedalus_recipe_dispatch_h264_idct4 instead of ff_h264_idct_add_neon.
++ * The recipe layer picks the substrate (CPU NEON by default for
++ * cycle 6; future cycles may dispatch to V3D opportunistically).
++ *
++ * FFmpeg's 4x4 block memory layout matches daedalus's column-major
++ * convention: block[r + 4*c] = coefficient at (row r, col c).  Both
++ * sides destructively zero the block after the transform.
++ *
++ * The library context is process-global and lazily initialised under
++ * pthread_once.  We pick the no-QPU constructor here because
++ * libavcodec.so is loaded into arbitrary host processes
++ * (firefox-fourier, mpv-fourier, daedalus_v4l2_daemon, ...) and we
++ * cannot assume the host has a usable Vulkan instance.  Higher cycles
++ * (deblock, MC) that benefit from the QPU initialise their own
++ * recipe-selected context once that path is wired.
++ */
++
++#include <pthread.h>
++#include <stddef.h>
++#include <stdint.h>
++
++#include <daedalus.h>
++
++#include "libavutil/attributes.h"
++#include "libavcodec/h264dsp.h"
++
++static daedalus_ctx     *g_dctx;
++static pthread_once_t    g_dctx_once = PTHREAD_ONCE_INIT;
++
++static void daedalus_ctx_init_once(void)
++{
++    g_dctx = daedalus_ctx_create_no_qpu();
++}
++
++void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride);
++
++void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
++{
++    static const daedalus_h264_block_meta meta = { .dst_off = 0 };
++
++    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
++
++    daedalus_recipe_dispatch_h264_idct4(g_dctx, dst, (size_t)stride,
++                                        block, 1, &meta);
++}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+index c684574..b993df2 100644
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c
++++ b/libavcodec/aarch64/h264dsp_init_aarch64.c
+@@ -66,6 +66,7 @@ void ff_biweight_h264_pixels_4_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride
+                                     int weights, int offset);
+ 
+ void ff_h264_idct_add_neon(uint8_t *dst, int16_t *block, int stride);
++void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride);
+ void ff_h264_idct_dc_add_neon(uint8_t *dst, int16_t *block, int stride);
+ void ff_h264_idct_add16_neon(uint8_t *dst, const int *block_offset,
+                              int16_t *block, int stride,
+@@ -139,7 +140,7 @@ av_cold void ff_h264dsp_init_aarch64(H264DSPContext *c, const int bit_depth,
+         c->biweight_pixels_tab[1] = ff_biweight_h264_pixels_8_neon;
+         c->biweight_pixels_tab[2] = ff_biweight_h264_pixels_4_neon;
+ 
+-        c->idct_add        = ff_h264_idct_add_neon;
++        c->idct_add        = ff_h264_idct_add_daedalus;
+         c->idct_dc_add     = ff_h264_idct_dc_add_neon;
+         c->idct_add16      = ff_h264_idct_add16_neon;
+         c->idct_add16intra = ff_h264_idct_add16intra_neon;
+-- 
+2.47.3
+

arch/ffmpeg-v4l2-request-fourier/PKGBUILD

@@ -24,8 +24,13 @@ _srcname=FFmpeg
 _version='8.1'
 _commit='b57fbbe50c9b2656fad86a1a7eeabfd2b2a50935'  # v4l2-request-n8.1 tip 2026-04-24
 pkgver=8.1.r123329.b57fbbe
-pkgrel=5
+pkgrel=6   # pkgrel=6 — H.264 IDCT 4x4 daedalus-fourier substitution (2026-05-21)
 epoch=2
+
+# daedalus-fourier pin — first kernel substitution in libavcodec
+# (cycle 6 H.264 IDCT 4x4).  Same SHA as the daedalus-v4l2 daemon's
+# inline build; lockstep with that until the public API rolls.
+_daedalus_fourier_commit='d87239d8172307d9a1b93c95cbed116d175b85cc'
 pkgdesc='FFmpeg with V4L2 Request API hwaccel (Rockchip / Allwinner stateless decode)'
 arch=('aarch64')
 url='https://github.com/Kwiboo/FFmpeg'
@@ -34,6 +39,7 @@ depends=(
   alsa-lib
   bzip2
   fontconfig
+  vulkan-icd-loader
   fribidi
   gmp
   gnutls
@@ -59,10 +65,13 @@ depends=(
   zlib
 )
 makedepends=(
+  cmake
   git
   linux-api-headers
   mesa
   nasm
+  ninja
+  vulkan-headers
 )
 provides=(
   libavcodec.so
@@ -78,9 +87,11 @@ provides=(
 conflicts=(ffmpeg)
 replaces=(ffmpeg ffmpeg-v4l2-request-git)
 source=("git+https://github.com/Kwiboo/FFmpeg.git#commit=${_commit}"
+        "daedalus-fourier-${_daedalus_fourier_commit}.tar.gz::https://git.reauktion.de/marfrit/daedalus-fourier/archive/${_daedalus_fourier_commit}.tar.gz"
         '0001-libudev-bypass-fallback.patch'
-        '0002-nv15-to-p010-unpack.patch')
-sha256sums=('SKIP' 'SKIP' 'SKIP')
+        '0002-nv15-to-p010-unpack.patch'
+        '0003-h264-idct4-daedalus-fourier.patch')
+sha256sums=('SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP')
 
 pkgver() {
   cd "${_srcname}"
@@ -93,9 +104,25 @@ prepare() {
   cd "${_srcname}"
   patch -Np1 -i "${srcdir}/0001-libudev-bypass-fallback.patch"
   patch -Np1 -i "${srcdir}/0002-nv15-to-p010-unpack.patch"
+  patch -Np1 -i "${srcdir}/0003-h264-idct4-daedalus-fourier.patch"
 }
 
 build() {
+  # --- daedalus-fourier: build static .a with PIC, install to a
+  # per-build prefix; libavcodec.so links it into the shared object so
+  # H264DSPContext.idct_add (and follow-up kernels) dispatch through
+  # the daedalus recipe layer instead of the in-tree NEON .S code. ---
+  local _fourier_prefix="${srcdir}/fourier-prefix"
+  mkdir -p "${_fourier_prefix}"
+  pushd "${srcdir}"/daedalus-fourier >/dev/null
+  cmake -B build -G Ninja \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+    -DCMAKE_INSTALL_PREFIX="${_fourier_prefix}"
+  cmake --build build --target daedalus_core
+  cmake --install build
+  popd >/dev/null
+
   cd "${_srcname}"
 
   # FFmpeg's configure resolves the compiler via `which` and bakes the
@@ -147,6 +174,9 @@ build() {
     --enable-libx265 \
     --enable-libwebp \
     \
+    --extra-cflags="-I${_fourier_prefix}/include" \
+    --extra-ldflags="-L${_fourier_prefix}/lib" \
+    --extra-libs="-ldaedalus_core -lvulkan -lpthread" \
     --host-cflags='-fPIC'
 
   make

arch/mesa-panvk-bifrost/0004-panvk-bifrost-xfb-primitive-decomposition.patch

@@ -0,0 +1,629 @@
+diff -urN a/src/panfrost/vulkan/meson.build b/src/panfrost/vulkan/meson.build
+--- a/src/panfrost/vulkan/meson.build	2026-05-21 14:04:02.529474145 +0200
++++ b/src/panfrost/vulkan/meson.build	2026-05-21 14:04:04.106755486 +0200
+@@ -123,6 +123,7 @@
+   'panvk_vX_nir_lower_input_attachment_loads.c',
+   'panvk_vX_sampler.c',
+   'panvk_vX_shader.c',
++  'panvk_vX_xfb_lower.c',
+   sha1_h,
+ ]
+ 
+diff -urN a/src/panfrost/vulkan/panvk_shader.h b/src/panfrost/vulkan/panvk_shader.h
+--- a/src/panfrost/vulkan/panvk_shader.h	2026-05-21 14:04:02.525251986 +0200
++++ b/src/panfrost/vulkan/panvk_shader.h	2026-05-21 14:04:04.084251800 +0200
+@@ -154,6 +154,8 @@
+       /* aligned_u64 attribute below inserts the 4-byte alignment gap
+        * after num_vertices automatically — no explicit pad needed. */
+       aligned_u64 xfb_address[4];  /* iter13: 4 transform feedback buffer base addresses */
++      uint32_t xfb_topology;       /* iter17: panvk_xfb_topology enum value */
++      uint32_t xfb_output_count;   /* iter17: per-instance output verts after decomp */
+ #endif
+       int32_t first_vertex;
+       int32_t base_instance;
+@@ -569,4 +571,76 @@
+    struct pan_compute_dim local_size, const void *bin_ptr, size_t bin_size,
+    struct panvk_shader **shader_out);
+ 
++
++#if PAN_ARCH < 9
++/* iter17: encoding for vs.xfb_topology sysval. Maps VkPrimitiveTopology values
++ * we need to distinguish at shader runtime for XFB capture. LIST topologies
++ * use the iter13 single-store fast path; non-LIST need per-vertex decomposition. */
++enum panvk_xfb_topology {
++   PANVK_XFB_TOPO_LIST            = 0,
++   PANVK_XFB_TOPO_LINE_STRIP      = 1,
++   PANVK_XFB_TOPO_TRI_STRIP       = 2,
++   PANVK_XFB_TOPO_TRI_FAN         = 3,
++   PANVK_XFB_TOPO_LINE_LIST_ADJ   = 4,
++   PANVK_XFB_TOPO_LINE_STRIP_ADJ  = 5,
++   PANVK_XFB_TOPO_TRI_LIST_ADJ    = 6,
++   PANVK_XFB_TOPO_TRI_STRIP_ADJ   = 7,
++};
++
++#include "panvk_macros.h"
++struct nir_shader;
++bool panvk_per_arch(nir_lower_xfb)(struct nir_shader *nir);
++
++/* Map VkPrimitiveTopology to panvk_xfb_topology enum (driver-side helper). */
++static inline uint32_t
++panvk_vk_topology_to_xfb_enum(VkPrimitiveTopology topo)
++{
++   switch (topo) {
++   case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP:
++      return PANVK_XFB_TOPO_LINE_STRIP;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP:
++      return PANVK_XFB_TOPO_TRI_STRIP;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN:
++      return PANVK_XFB_TOPO_TRI_FAN;
++   case VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY:
++      return PANVK_XFB_TOPO_LINE_LIST_ADJ;
++   case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY:
++      return PANVK_XFB_TOPO_LINE_STRIP_ADJ;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY:
++      return PANVK_XFB_TOPO_TRI_LIST_ADJ;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY:
++      return PANVK_XFB_TOPO_TRI_STRIP_ADJ;
++   case VK_PRIMITIVE_TOPOLOGY_POINT_LIST:
++   case VK_PRIMITIVE_TOPOLOGY_LINE_LIST:
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST:
++   default:
++      return PANVK_XFB_TOPO_LIST;
++   }
++}
++
++/* Compute the per-instance output vertex count for a given (topology, input count). */
++static inline uint32_t
++panvk_xfb_output_count(VkPrimitiveTopology topo, uint32_t input_count)
++{
++   switch (topo) {
++   case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP:
++      return input_count >= 1 ? 2u * (input_count - 1u) : 0u;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP:
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN:
++      return input_count >= 2 ? 3u * (input_count - 2u) : 0u;
++   case VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY:
++      return (input_count / 4u) * 2u;
++   case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY:
++      return input_count >= 3 ? 2u * (input_count - 3u) : 0u;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY:
++      return (input_count / 6u) * 3u;
++   case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY:
++      return input_count >= 6 ? 3u * (input_count / 2u - 2u) : 0u;
++   default:
++      return input_count;  /* LIST topologies: 1:1 mapping */
++   }
++}
++#endif
++
++
+ #endif
+diff -urN a/src/panfrost/vulkan/panvk_vX_cmd_draw.c b/src/panfrost/vulkan/panvk_vX_cmd_draw.c
+--- a/src/panfrost/vulkan/panvk_vX_cmd_draw.c	2026-05-21 14:04:02.528576354 +0200
++++ b/src/panfrost/vulkan/panvk_vX_cmd_draw.c	2026-05-21 14:04:04.091357598 +0200
+@@ -727,6 +727,20 @@
+    /* iter13: VK_EXT_transform_feedback sysvals — always set (per draw),
+     * reflect bound XFB state. set_gfx_sysval is a no-op if value unchanged. */
+    set_gfx_sysval(cmdbuf, dirty_sysvals, vs.num_vertices, info->vertex.count);
++
++   /* iter17: XFB primitive-decomposition sysvals.
++    * xfb_topology = enum value for the current bound topology.
++    * xfb_output_count = per-instance output vertex count after decomposition.
++    * For LIST topologies, output_count == input vertex count and the shader
++    * takes the iter13 single-store fast path. */
++   {
++      VkPrimitiveTopology vk_topo =
++         cmdbuf->vk.dynamic_graphics_state.ia.primitive_topology;
++      uint32_t topo_enum = panvk_vk_topology_to_xfb_enum(vk_topo);
++      uint32_t out_count = panvk_xfb_output_count(vk_topo, info->vertex.count);
++      set_gfx_sysval(cmdbuf, dirty_sysvals, vs.xfb_topology, topo_enum);
++      set_gfx_sysval(cmdbuf, dirty_sysvals, vs.xfb_output_count, out_count);
++   }
+    {
+       const struct panvk_cmd_graphics_state *_gfx = &cmdbuf->state.gfx;
+       /* iter13: default each XFB buffer address to PAN_SHADER_OOB_ADDRESS
+diff -urN a/src/panfrost/vulkan/panvk_vX_shader.c b/src/panfrost/vulkan/panvk_vX_shader.c
+--- a/src/panfrost/vulkan/panvk_vX_shader.c	2026-05-21 14:04:02.527576494 +0200
++++ b/src/panfrost/vulkan/panvk_vX_shader.c	2026-05-21 14:04:04.098356619 +0200
+@@ -895,7 +895,10 @@
+        nir->info.has_transform_feedback_varyings) {
+       NIR_PASS(_, nir, nir_opt_constant_folding);
+       NIR_PASS(_, nir, nir_io_add_intrinsic_xfb_info);
+-      NIR_PASS(_, nir, pan_nir_lower_xfb);
++      /* iter17: panvk-specific replacement for pan_nir_lower_xfb that handles
++       * primitive decomposition for non-LIST topologies. Single-store LIST
++       * fast path matches iter13 behavior. */
++      NIR_PASS(_, nir, panvk_per_arch(nir_lower_xfb));
+    }
+ #endif
+ }
+diff -urN a/src/panfrost/vulkan/panvk_vX_xfb_lower.c b/src/panfrost/vulkan/panvk_vX_xfb_lower.c
+--- a/src/panfrost/vulkan/panvk_vX_xfb_lower.c	1970-01-01 01:00:00.000000000 +0100
++++ b/src/panfrost/vulkan/panvk_vX_xfb_lower.c	2026-05-21 14:04:04.115354242 +0200
+@@ -0,0 +1,486 @@
++/*
++ * Copyright © 2026 mfritsche / claude-noether
++ * SPDX-License-Identifier: MIT
++ *
++ * iter17: panvk-specific replacement for pan_nir_lower_xfb that handles
++ * primitive decomposition for transform_feedback on non-LIST topologies
++ * (TRIANGLE_STRIP/FAN, LINE_STRIP, *_WITH_ADJACENCY).
++ *
++ * Approach: emit a topology dispatch at the start of each store_output
++ * lowering. The shader reads vs.xfb_topology sysval at runtime and branches
++ * into per-topology emission logic. For each affected topology, the lowered
++ * code emits guarded conditional stores — one per primitive this vertex
++ * contributes to, computing the output buffer position via primitive index
++ * and slot within the decomposed primitive.
++ *
++ * For LIST topologies (POINT/LINE/TRIANGLE LIST), takes a fast path that
++ * matches iter13's single-store behavior.
++ *
++ * For TRIANGLE_FAN, the central vertex (v=0) contributes to ALL primitives
++ * as slot 2 — handled via a NIR loop bounded by num_vertices.
++ *
++ * See ~/src/panvk-bifrost/iter17/phase{0,1,2}_*.md for full design context.
++ */
++
++#include "panvk_macros.h"
++
++#if PAN_ARCH < 9
++
++#include "panvk_shader.h"
++
++#include "compiler/nir/nir_builder.h"
++#include "pan_nir.h"
++
++#include <vulkan/vulkan_core.h>
++
++/* ----- Address arithmetic ----- */
++
++static nir_def *
++xfb_store_addr(nir_builder *b, nir_def *buf, nir_def *out_idx,
++               uint16_t stride, uint16_t offset_bytes)
++{
++   nir_def *byte_off = nir_iadd_imm(b,
++      nir_imul_imm(b, out_idx, stride), offset_bytes);
++   return nir_iadd(b, buf, nir_u2u64(b, byte_off));
++}
++
++static void
++emit_list_store(nir_builder *b, nir_def *buf, nir_def *output_count,
++                nir_def *instance_id, nir_def *raw_vid, nir_def *value,
++                uint16_t stride, uint16_t offset_bytes)
++{
++   nir_def *out_idx = nir_iadd(b,
++      nir_imul(b, instance_id, output_count), raw_vid);
++   nir_def *addr = xfb_store_addr(b, buf, out_idx, stride, offset_bytes);
++   nir_store_global(b, value, addr);
++}
++
++static void
++emit_prim_store(nir_builder *b, nir_def *buf, nir_def *output_count,
++                nir_def *instance_id, nir_def *eligible,
++                nir_def *prim_idx, nir_def *slot,
++                uint32_t verts_per_prim,
++                nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   nir_push_if(b, eligible);
++   {
++      nir_def *out_idx = nir_iadd(b,
++         nir_imul(b, instance_id, output_count),
++         nir_iadd(b, nir_imul_imm(b, prim_idx, verts_per_prim), slot));
++      nir_def *addr = xfb_store_addr(b, buf, out_idx, stride, offset_bytes);
++      nir_store_global(b, value, addr);
++   }
++   nir_pop_if(b, NULL);
++}
++
++/* ----- Per-topology emission ----- */
++
++/* TRIANGLE_STRIP: vertex v contributes to prims v, v-1, v-2 (per eligibility). */
++static void
++emit_tri_strip(nir_builder *b, nir_def *v, nir_def *N,
++               nir_def *buf, nir_def *output_count, nir_def *instance_id,
++               nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   nir_def *Nm2 = nir_iadd_imm(b, N, -2);
++   nir_def *Nm1 = nir_iadd_imm(b, N, -1);
++
++   /* Prim v, slot 0: v < N-2 */
++   emit_prim_store(b, buf, output_count, instance_id,
++      nir_ult(b, v, Nm2),
++      v, nir_imm_int(b, 0), 3, value, stride, offset_bytes);
++
++   /* Prim v-1, slot = 1 if prim even else 2: 1 <= v < N-1 */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -1);
++      nir_def *parity = nir_iand_imm(b, prim, 1u);
++      nir_def *slot = nir_iadd_imm(b, parity, 1);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 1)),
++         nir_ult(b, v, Nm1));
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, slot, 3, value, stride, offset_bytes);
++   }
++
++   /* Prim v-2, slot = 2 if prim even else 1: 2 <= v < N */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -2);
++      nir_def *parity = nir_iand_imm(b, prim, 1u);
++      nir_def *slot = nir_isub(b, nir_imm_int(b, 2), parity);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 2)),
++         nir_ult(b, v, N));
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, slot, 3, value, stride, offset_bytes);
++   }
++}
++
++/* LINE_STRIP: vertex v contributes to prim v slot 0 + prim v-1 slot 1. */
++static void
++emit_line_strip(nir_builder *b, nir_def *v, nir_def *N,
++                nir_def *buf, nir_def *output_count, nir_def *instance_id,
++                nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   nir_def *Nm1 = nir_iadd_imm(b, N, -1);
++
++   /* Prim v, slot 0: v < N-1 */
++   emit_prim_store(b, buf, output_count, instance_id,
++      nir_ult(b, v, Nm1),
++      v, nir_imm_int(b, 0), 2, value, stride, offset_bytes);
++
++   /* Prim v-1, slot 1: 1 <= v < N */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -1);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 1)),
++         nir_ult(b, v, N));
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, nir_imm_int(b, 1), 2, value, stride, offset_bytes);
++   }
++}
++
++/* TRIANGLE_FAN: prim p emits {p+1, p+2, 0}.
++ *   vertex v=0: contributes to ALL prims as slot 2 (loop required)
++ *   vertex v>=1: contributes to prim v-1 as slot 0 (if 1 <= v <= N-2)
++ *   vertex v>=2: contributes to prim v-2 as slot 1 (if 2 <= v <= N-1)
++ */
++static void
++emit_tri_fan(nir_builder *b, nir_def *v, nir_def *N,
++             nir_def *buf, nir_def *output_count, nir_def *instance_id,
++             nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   nir_def *Nm1 = nir_iadd_imm(b, N, -1);
++   nir_def *Nm2 = nir_iadd_imm(b, N, -2);
++
++   /* Prim v-1, slot 0: 1 <= v < N-1 */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -1);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 1)),
++         nir_ult(b, v, Nm1));
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, nir_imm_int(b, 0), 3, value, stride, offset_bytes);
++   }
++
++   /* Prim v-2, slot 1: 2 <= v < N */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -2);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 2)),
++         nir_ult(b, v, N));
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, nir_imm_int(b, 1), 3, value, stride, offset_bytes);
++   }
++
++   /* Central vertex (v == 0): loop over all prims, write to slot 2. */
++   nir_push_if(b, nir_ieq_imm(b, v, 0));
++   {
++      nir_variable *p_var = nir_local_variable_create(b->impl,
++         glsl_uint_type(), "fan_p");
++      nir_store_var(b, p_var, nir_imm_int(b, 0), 0x1);
++      nir_push_loop(b);
++      {
++         nir_def *p = nir_load_var(b, p_var);
++         nir_push_if(b, nir_uge(b, p, Nm2));
++         {
++            nir_jump(b, nir_jump_break);
++         }
++         nir_pop_if(b, NULL);
++
++         nir_def *out_idx = nir_iadd(b,
++            nir_imul(b, instance_id, output_count),
++            nir_iadd_imm(b, nir_imul_imm(b, p, 3), 2));
++         nir_def *addr = xfb_store_addr(b, buf, out_idx, stride, offset_bytes);
++         nir_store_global(b, value, addr);
++
++         nir_store_var(b, p_var, nir_iadd_imm(b, p, 1), 0x1);
++      }
++      nir_pop_loop(b, NULL);
++   }
++   nir_pop_if(b, NULL);
++}
++
++/* LINE_LIST_WITH_ADJACENCY: 4-vertex groups [4i..4i+3]; output {4i+1, 4i+2}.
++ *   v contributes if v%4 == 1: prim v/4 slot 0
++ *   v contributes if v%4 == 2: prim v/4 slot 1
++ */
++static void
++emit_line_list_adj(nir_builder *b, nir_def *v, nir_def *N,
++                   nir_def *buf, nir_def *output_count, nir_def *instance_id,
++                   nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   (void)N; /* eligibility is mod-based, not range-based */
++   nir_def *vmod4 = nir_iand_imm(b, v, 3u);
++   nir_def *prim = nir_ushr_imm(b, v, 2);  /* v / 4 */
++
++   emit_prim_store(b, buf, output_count, instance_id,
++      nir_ieq_imm(b, vmod4, 1),
++      prim, nir_imm_int(b, 0), 2, value, stride, offset_bytes);
++
++   emit_prim_store(b, buf, output_count, instance_id,
++      nir_ieq_imm(b, vmod4, 2),
++      prim, nir_imm_int(b, 1), 2, value, stride, offset_bytes);
++}
++
++/* LINE_STRIP_WITH_ADJACENCY: prim p emits {p+1, p+2}.
++ *   v contributes to prim v-1 slot 0 (1 <= v <= N-2)
++ *   v contributes to prim v-2 slot 1 (2 <= v <= N-1)
++ */
++static void
++emit_line_strip_adj(nir_builder *b, nir_def *v, nir_def *N,
++                    nir_def *buf, nir_def *output_count, nir_def *instance_id,
++                    nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   nir_def *Nm1 = nir_iadd_imm(b, N, -1);
++   nir_def *Nm2 = nir_iadd_imm(b, N, -2);
++
++   /* Prim v-1, slot 0: 1 <= v <= N-2 ⇔ v >= 1 AND v <= N-2 ⇔ v >= 1 AND v < N-1 */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -1);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 1)),
++         nir_ult(b, v, Nm1));
++      (void)Nm2;
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, nir_imm_int(b, 0), 2, value, stride, offset_bytes);
++   }
++
++   /* Prim v-2, slot 1: 2 <= v <= N-1 ⇔ v >= 2 AND v < N */
++   {
++      nir_def *prim = nir_iadd_imm(b, v, -2);
++      nir_def *eligible = nir_iand(b,
++         nir_uge(b, v, nir_imm_int(b, 2)),
++         nir_ult(b, v, N));
++      emit_prim_store(b, buf, output_count, instance_id, eligible,
++                      prim, nir_imm_int(b, 1), 2, value, stride, offset_bytes);
++   }
++}
++
++/* TRIANGLE_LIST_WITH_ADJACENCY: 6-vertex groups; output {6i, 6i+2, 6i+4}.
++ *   v contributes if v%6 == 0: prim v/6 slot 0
++ *   v contributes if v%6 == 2: prim v/6 slot 1
++ *   v contributes if v%6 == 4: prim v/6 slot 2
++ */
++static void
++emit_tri_list_adj(nir_builder *b, nir_def *v, nir_def *N,
++                  nir_def *buf, nir_def *output_count, nir_def *instance_id,
++                  nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   (void)N;
++   nir_def *vmod6 = nir_umod_imm(b, v, 6);
++   nir_def *prim = nir_udiv_imm(b, v, 6);
++
++   for (uint32_t slot = 0; slot < 3; slot++) {
++      emit_prim_store(b, buf, output_count, instance_id,
++         nir_ieq_imm(b, vmod6, slot * 2),
++         prim, nir_imm_int(b, slot), 3, value, stride, offset_bytes);
++   }
++}
++
++/* TRIANGLE_STRIP_WITH_ADJACENCY: prim i emits:
++ *   even i: {2i, 2i+2, 2i+4}    (slots 0, 1, 2 ← input indices 2i, 2i+2, 2i+4)
++ *   odd  i: {2i, 2i+4, 2i+2}    (slots 0, 1, 2 ← input indices 2i, 2i+4, 2i+2)
++ *
++ * Only EVEN input vertices contribute (since all output indices are 2*something).
++ * For even input v:
++ *   prim v/2 slot 0 (always, if v/2 < N/2-2)
++ *   prim (v-2)/2 slot 1 if (v-2)/2 even, slot 2 if odd   (when v >= 2)
++ *   prim (v-4)/2 slot 2 if (v-4)/2 even, slot 1 if odd   (when v >= 4)
++ */
++static void
++emit_tri_strip_adj(nir_builder *b, nir_def *v, nir_def *N,
++                   nir_def *buf, nir_def *output_count, nir_def *instance_id,
++                   nir_def *value, uint16_t stride, uint16_t offset_bytes)
++{
++   /* Bail for odd input vertices — they never contribute. */
++   nir_def *v_is_even = nir_ieq_imm(b, nir_iand_imm(b, v, 1u), 0);
++   nir_push_if(b, v_is_even);
++   {
++      nir_def *N_half = nir_ushr_imm(b, N, 1);
++      nir_def *max_prim = nir_iadd_imm(b, N_half, -2);  /* N/2 - 2 */
++      nir_def *v_half = nir_ushr_imm(b, v, 1);
++
++      /* Prim v/2 slot 0: v/2 < N/2 - 2 */
++      emit_prim_store(b, buf, output_count, instance_id,
++         nir_ult(b, v_half, max_prim),
++         v_half, nir_imm_int(b, 0), 3, value, stride, offset_bytes);
++
++      /* Prim (v-2)/2 = v/2 - 1: v >= 2 AND prim < N/2-2 */
++      {
++         nir_def *prim = nir_iadd_imm(b, v_half, -1);
++         nir_def *parity = nir_iand_imm(b, prim, 1u);
++         nir_def *slot = nir_iadd_imm(b, parity, 1);  /* even→1, odd→2 */
++         nir_def *eligible = nir_iand(b,
++            nir_uge(b, v, nir_imm_int(b, 2)),
++            nir_ult(b, prim, max_prim));
++         emit_prim_store(b, buf, output_count, instance_id, eligible,
++                         prim, slot, 3, value, stride, offset_bytes);
++      }
++
++      /* Prim (v-4)/2 = v/2 - 2: v >= 4 AND prim < N/2-2 */
++      {
++         nir_def *prim = nir_iadd_imm(b, v_half, -2);
++         nir_def *parity = nir_iand_imm(b, prim, 1u);
++         nir_def *slot = nir_isub(b, nir_imm_int(b, 2), parity);  /* even→2, odd→1 */
++         nir_def *eligible = nir_iand(b,
++            nir_uge(b, v, nir_imm_int(b, 4)),
++            nir_ult(b, prim, max_prim));
++         emit_prim_store(b, buf, output_count, instance_id, eligible,
++                         prim, slot, 3, value, stride, offset_bytes);
++      }
++   }
++   nir_pop_if(b, NULL);
++}
++
++/* ----- Main lowering: per store_output XFB channel ----- */
++
++static void
++lower_xfb_output_iter17(nir_builder *b, nir_intrinsic_instr *intr,
++                        unsigned channel_idx, unsigned num_components,
++                        unsigned buffer, unsigned offset_words)
++{
++   assert(buffer < MAX_XFB_BUFFERS);
++   assert(nir_intrinsic_component(intr) == 0);
++
++   uint16_t stride = b->shader->info.xfb_stride[buffer] * 4;
++   assert(stride != 0);
++   uint16_t offset_bytes = offset_words * 4;
++
++   BITSET_SET(b->shader->info.system_values_read, SYSTEM_VALUE_VERTEX_ID_ZERO_BASE);
++   BITSET_SET(b->shader->info.system_values_read, SYSTEM_VALUE_INSTANCE_ID);
++
++   nir_def *topology = load_sysval(b, graphics, 32, vs.xfb_topology);
++   nir_def *out_count = load_sysval(b, graphics, 32, vs.xfb_output_count);
++   nir_def *N = nir_load_num_vertices(b);
++   nir_def *v = nir_load_raw_vertex_id_pan(b);
++   nir_def *instance = nir_load_instance_id(b);
++   nir_def *buf = nir_load_xfb_address(b, 64, .base = buffer);
++
++   nir_def *src = intr->src[0].ssa;
++   nir_component_mask_t mask = nir_component_mask(num_components);
++   nir_def *value = nir_channels(b, src, mask << channel_idx);
++
++   /* Topology dispatch ladder. LIST first (fast path). */
++   nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_LIST));
++   {
++      emit_list_store(b, buf, out_count, instance, v, value,
++                      stride, offset_bytes);
++   }
++   nir_push_else(b, NULL);
++   {
++      /* iter17 Janet Finding 3: gate all non-LIST emission on
++       * output_count > 0. For degenerate input counts (N < min required
++       * for the topology), output_count is 0 and we must emit NO stores
++       * — otherwise N-2 / N-3 / etc. arithmetic underflows in the
++       * eligibility predicates and we falsely fire stores. */
++      nir_push_if(b, nir_ult(b, nir_imm_int(b, 0), out_count));
++      {
++      nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_TRI_STRIP));
++      {
++         emit_tri_strip(b, v, N, buf, out_count, instance, value,
++                        stride, offset_bytes);
++      }
++      nir_push_else(b, NULL);
++      {
++         nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_LINE_STRIP));
++         {
++            emit_line_strip(b, v, N, buf, out_count, instance, value,
++                            stride, offset_bytes);
++         }
++         nir_push_else(b, NULL);
++         {
++            nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_TRI_FAN));
++            {
++               emit_tri_fan(b, v, N, buf, out_count, instance, value,
++                            stride, offset_bytes);
++            }
++            nir_push_else(b, NULL);
++            {
++               nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_LINE_LIST_ADJ));
++               {
++                  emit_line_list_adj(b, v, N, buf, out_count, instance, value,
++                                     stride, offset_bytes);
++               }
++               nir_push_else(b, NULL);
++               {
++                  nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_LINE_STRIP_ADJ));
++                  {
++                     emit_line_strip_adj(b, v, N, buf, out_count, instance, value,
++                                         stride, offset_bytes);
++                  }
++                  nir_push_else(b, NULL);
++                  {
++                     nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_TRI_LIST_ADJ));
++                     {
++                        emit_tri_list_adj(b, v, N, buf, out_count, instance, value,
++                                          stride, offset_bytes);
++                     }
++                     nir_push_else(b, NULL);
++                     {
++                        /* TRI_STRIP_ADJ — last case */
++                        emit_tri_strip_adj(b, v, N, buf, out_count, instance, value,
++                                           stride, offset_bytes);
++                     }
++                     nir_pop_if(b, NULL);
++                  }
++                  nir_pop_if(b, NULL);
++               }
++               nir_pop_if(b, NULL);
++            }
++            nir_pop_if(b, NULL);
++         }
++         nir_pop_if(b, NULL);
++      }
++      nir_pop_if(b, NULL);
++      }
++      nir_pop_if(b, NULL);  /* Janet Finding 3: close output_count > 0 guard */
++   }
++   nir_pop_if(b, NULL);
++}
++
++/* Mirror of pan_nir_lower_xfb's lower_xfb: load_vertex_id rewrite +
++ * dispatch store_output through our topology-aware emission. */
++static bool
++lower_xfb_iter17(nir_builder *b, nir_intrinsic_instr *intr,
++                 UNUSED void *data)
++{
++   if (intr->intrinsic == nir_intrinsic_load_vertex_id) {
++      b->cursor = nir_instr_remove(&intr->instr);
++      nir_def *repl = nir_iadd(b, nir_load_raw_vertex_id_pan(b),
++                               nir_load_raw_vertex_offset_pan(b));
++      nir_def_rewrite_uses(&intr->def, repl);
++      return true;
++   }
++
++   if (intr->intrinsic != nir_intrinsic_store_output)
++      return false;
++
++   bool progress = false;
++   b->cursor = nir_before_instr(&intr->instr);
++
++   /* io_xfb has only out[0,1]; the other 2 channels are in io_xfb2.
++    * Outer loop selects which annotation; inner picks which channel. */
++   for (unsigned i = 0; i < 2; ++i) {
++      nir_io_xfb xfb = i ? nir_intrinsic_io_xfb2(intr)
++                         : nir_intrinsic_io_xfb(intr);
++      for (unsigned j = 0; j < 2; ++j) {
++         if (!xfb.out[j].num_components)
++            continue;
++         lower_xfb_output_iter17(b, intr, i * 2 + j, xfb.out[j].num_components,
++                                 xfb.out[j].buffer, xfb.out[j].offset);
++         progress = true;
++      }
++   }
++
++   if (progress)
++      nir_instr_remove(&intr->instr);
++   return progress;
++}
++
++bool
++panvk_per_arch(nir_lower_xfb)(nir_shader *nir)
++{
++   return nir_shader_intrinsics_pass(
++      nir, lower_xfb_iter17, nir_metadata_control_flow, NULL);
++}
++
++#endif /* PAN_ARCH < 9 */

arch/mesa-panvk-bifrost/PKGBUILD

@@ -30,7 +30,7 @@
 
 pkgname=mesa-panvk-bifrost
 _mesaver=26.0.6
-pkgver=26.0.6.r3
+pkgver=26.0.6.r4
 pkgrel=1
 pkgdesc="Patched Mesa libvulkan_panfrost.so exposing Bifrost-gen Mali to Vulkan apps (panvk-bifrost campaign)"
 arch=('aarch64')
@@ -80,6 +80,7 @@ source=(
     "0001-panvk-expose-robustness2-nullDescriptor-bifrost.patch"
     "0002-panvk-expose-vulkan-1.1-1.2-on-bifrost.patch"
     "0003-panvk-bifrost-vk-ext-transform-feedback.patch"
+    "0004-panvk-bifrost-xfb-primitive-decomposition.patch"
     "brave-vulkan"
     "icd.json"
 )
@@ -90,6 +91,7 @@ sha256sums=(
     'SKIP'
     'SKIP'
     'SKIP'
+    'SKIP'
 )
 
 prepare() {
@@ -116,6 +118,15 @@ prepare() {
     # reports "Hardware accelerated" across the board for the affected paths).
     patch -p1 < "${srcdir}/0003-panvk-bifrost-vk-ext-transform-feedback.patch"
 
+    # iter17: XFB primitive decomposition for non-LIST topologies (TRI_STRIP,
+    # TRI_FAN, LINE_STRIP, *_WITH_ADJACENCY). Replacement panvk-specific
+    # NIR pass (panvk_per_arch(nir_lower_xfb)) substituted for upstream
+    # pan_nir_lower_xfb. Closes the 162 dEQP-VK winding_* failures from
+    # iter15 (958 P / 81 F / 0 Crash on full XFB CTS — remaining 81 fails
+    # are by-design resume_* tests, transformFeedbackDraw=false).
+    # Phase-doc context: ~/src/panvk-bifrost/iter17/phase{0,1,2,4,5,6,8}_*.md.
+    patch -p1 < "${srcdir}/0004-panvk-bifrost-xfb-primitive-decomposition.patch"
+
     # Sanity-check the patches landed.
     grep -q "KHR_robustness2 = true," src/panfrost/vulkan/panvk_vX_physical_device.c
     grep -q "EXT_robustness2 = true," src/panfrost/vulkan/panvk_vX_physical_device.c
@@ -124,8 +135,12 @@ prepare() {
     grep -q "has_vk1_2 = true;" src/panfrost/vulkan/panvk_vX_physical_device.c
     # iter13 sanity:
     grep -q "EXT_transform_feedback = PAN_ARCH < 9," src/panfrost/vulkan/panvk_vX_physical_device.c
-    grep -q "pan_nir_lower_xfb" src/panfrost/vulkan/panvk_vX_shader.c
     test -f src/panfrost/vulkan/jm/panvk_vX_cmd_xfb.c
+    # iter17 sanity: pan_nir_lower_xfb call site has been replaced; new file present.
+    grep -q "panvk_per_arch(nir_lower_xfb)" src/panfrost/vulkan/panvk_vX_shader.c
+    grep -q "xfb_topology" src/panfrost/vulkan/panvk_shader.h
+    grep -q "panvk_xfb_topology" src/panfrost/vulkan/panvk_shader.h
+    test -f src/panfrost/vulkan/panvk_vX_xfb_lower.c
 }
 
 build() {

debian/daedalus-v4l2-dkms/build-deb.sh

@@ -14,9 +14,9 @@
 # Sibling userspace package: ../daedalus-v4l2/build-deb.sh
 set -euo pipefail
 
-UPSTREAM_COMMIT=79256dc7ef41f83873ca9c23db20f5888858e65d
-PKGVER=0.1.0+r28+g79256dc
-PKGREL=1  # reset for new upstream pin (79256dc — H.264 B-frame reorder fix); still carries the #64 multi-kernel postinst fix
+UPSTREAM_COMMIT=5d8b4369e58ab947d1c56b1f718293c57c6065b5
+PKGVER=0.1.0+r33+g5d8b436
+PKGREL=1  # reset for new upstream pin (5d8b436 — revert parking design); still carries the #64 multi-kernel postinst fix
 MODULE_NAME=daedalus_v4l2
 
 HERE=$(dirname "$(readlink -f "$0")")

debian/daedalus-v4l2-dkms/debian/changelog

@@ -1,3 +1,39 @@
+daedalus-v4l2-dkms (0.1.0+r33+g5d8b436-1) bookworm trixie; urgency=medium
+
+  * Bump to 5d8b436 — reverts daedalus-v4l2 PRs #7 + #8.  Kernel
+    module returns to the pre-#7 buf_done_and_job_finish completion
+    model: no src/dst lifecycle decoupling, no parked dst_bufs, no
+    1:1-contract violation against libva-v4l2-request-fourier
+    (closes daedalus-v4l2#9 + #10 as won't-fix at this layer; proper
+    fix tracked at daedalus-v4l2#11).
+  * Wire-protocol drops 1 → 0; lock-step install with daedalus-v4l2
+    0.1.0+r33+g5d8b436 REQUIRED.
+  * Carries forward the #64 multi-kernel postinst fix.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 14:50:00 +0000
+
+daedalus-v4l2-dkms (0.1.0+r30+g6ffe92b-1) bookworm trixie; urgency=medium
+
+  * Bump to 6ffe92b — fixes the kernel panic regression introduced
+    by 79256dc's split-completion design (closes daedalus-v4l2#8).
+    `device_run` now removes both src + dst from `m2m_ctx`'s
+    rdy_queue at pickup time, not at `buf_done` time.  Without
+    this, after `SRC_CONSUMED`'s `job_finish` released the m2m
+    scheduler, the NEXT `device_run` saw the still-queued parked
+    dst_buf and paired it with a fresh src — two inflight entries
+    referencing the same vb2_buffer, the later `HAS_PIXELS`
+    triggered list_del on an already-detached list_head, smashing
+    the rdy_queue → hard reboot on Pi CM5 during `mpv vaapi-copy`
+    playback of 720p H.264 (2026-05-21).
+  * Wire protocol unchanged — DAEDALUS_PROTO_VERSION stays at 1.
+    Daemon (userspace daedalus-v4l2 package) need NOT bump in
+    lockstep with this DKMS update; the existing
+    daedalus-v4l2 0.1.0+r28+g79256dc is wire-compatible with
+    daedalus-v4l2-dkms 0.1.0+r30+g6ffe92b.
+  * Carries forward the #64 multi-kernel postinst fix.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 14:00:00 +0000
+
 daedalus-v4l2-dkms (0.1.0+r28+g79256dc-1) bookworm trixie; urgency=medium
 
   * Bump to 79256dc — H.264 B-frame display reorder fix (closes

debian/daedalus-v4l2/build-deb.sh

@@ -11,16 +11,23 @@
 # Upstream repo: https://git.reauktion.de/reauktion/daedalus-v4l2
 set -euo pipefail
 
-# Same pin as the Arch PKGBUILD.  79256dc = "kernel + daemon: H.264
-# B-frame display reorder fix (closes #6)" — adds the wire-protocol
-# src_pts / output_src_pts / RESP_FRAME flags split that lets H.264
-# streams with B-frames preserve display order through libva → kernel
-# → daemon.  PROTO_VERSION bumps 0 → 1; lock-step userspace + kernel
-# rebuild REQUIRED (daedalus-v4l2-dkms build-deb.sh pinned to the same
-# commit).
-UPSTREAM_COMMIT=79256dc7ef41f83873ca9c23db20f5888858e65d
-PKGVER=0.1.0+r28+g79256dc
-PKGREL=1  # reset for new upstream pin (79256dc — H.264 B-frame reorder fix)
+# 6e6dfa1 = picks up daedalus-v4l2 PR #16 — daemon now dlopens
+# the Kwiboo fourier fork's libavcodec.so.62 / libavformat.so.62 /
+# libavutil.so.60 at /opt/fourier instead of Debian-stock soname
+# 61/61/59.  First step on the daedalus-fourier substitution arc
+# (daedalus-v4l2#11): routes the daemon through the libavcodec
+# source tree we own in marfrit-packages.  Headers + .pc files
+# come from ffmpeg-v4l2-request-fourier (installed by the CI
+# workflow before this script runs; see PKG_CONFIG_PATH below).
+UPSTREAM_COMMIT=6e6dfa144da7bc7fa8be50c8da91d7d1c6132a2c
+PKGVER=0.1.0+r41+g6e6dfa1
+PKGREL=1  # reset for new upstream pin (6e6dfa1 — soname 62 via /opt/fourier)
+
+# daedalus-fourier pin.  d87239d = marfrit/daedalus-fourier PR #1 merge
+# (install rules + pkg-config, enables this consumer to find_package
+# + link).  Bump in lockstep with the upstream daemon when daedalus-
+# fourier's API or installed shaders are changed by a new consumer.
+DAEDALUS_FOURIER_COMMIT=d87239d8172307d9a1b93c95cbed116d175b85cc
 
 HERE=$(dirname "$(readlink -f "$0")")
 
@@ -30,14 +37,37 @@ export SOURCE_DATE_EPOCH=1779231600
 work=$(mktemp -d)
 trap "rm -rf $work" EXIT
 
+# --- daedalus-fourier: fetch + build + install to per-build prefix ---
+#
+# Static-linked into the daemon, so the temp prefix is only for the
+# duration of this build script.  Requires libvulkan-dev + glslang-tools
+# on the runner (already needed for the daedalus-fourier benches).
+FOURIER_PREFIX=$work/fourier-prefix
+mkdir -p "$FOURIER_PREFIX"
+
+cd "$work"
+curl --connect-timeout 10 --max-time 600 --retry 3 --retry-delay 5 -sSLfo daedalus-fourier.tar.gz \
+    "https://git.reauktion.de/marfrit/daedalus-fourier/archive/${DAEDALUS_FOURIER_COMMIT}.tar.gz"
+tar xzf daedalus-fourier.tar.gz
+cd daedalus-fourier
+cmake -B build -G Ninja \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_INSTALL_PREFIX="$FOURIER_PREFIX"
+cmake --build build --target daedalus_core
+cmake --install build
+
+# --- daedalus-v4l2: fetch + build daemon against installed daedalus-fourier ---
+
 cd "$work"
 curl --connect-timeout 10 --max-time 600 --retry 3 --retry-delay 5 -sSLfo daedalus-v4l2.tar.gz \
     "https://git.reauktion.de/reauktion/daedalus-v4l2/archive/${UPSTREAM_COMMIT}.tar.gz"
 tar xzf daedalus-v4l2.tar.gz
 SRCDIR=daedalus-v4l2
 
-# Build daemon (CMake)
+# Build daemon (CMake) — point pkg-config at the daedalus-fourier
+# temp prefix so pkg_check_modules(DAEDALUS_FOURIER …) resolves to it.
 cd "$SRCDIR/daemon"
+PKG_CONFIG_PATH="$FOURIER_PREFIX/lib/pkgconfig:/opt/fourier/lib/pkgconfig" \
 cmake -B build -G Ninja \
     -DCMAKE_BUILD_TYPE=Release \
     -DCMAKE_INSTALL_PREFIX=/usr

debian/daedalus-v4l2/debian/changelog

@@ -1,3 +1,96 @@
+daedalus-v4l2 (0.1.0+r41+g6e6dfa1-1) bookworm trixie; urgency=medium
+
+  * Bump to 6e6dfa1 — daedalus-v4l2 PR #16.  Daemon dlopens Kwiboo
+    fourier fork's libavcodec.so.62 / libavformat.so.62 /
+    libavutil.so.60 at /opt/fourier instead of Debian-stock
+    soname 61/61/59.  First step on the daedalus-fourier
+    substitution arc (daedalus-v4l2#11): the next PR series
+    layers daedalus_recipe_dispatch_h264_* substitution patches
+    into ffmpeg-v4l2-request-fourier's H264DSPContext NEON init,
+    reaching the daemon's production decode path.
+  * Build: PKG_CONFIG_PATH now includes /opt/fourier/lib/pkgconfig
+    so daemon's pkg_check_modules picks up the Kwiboo .pc files.
+  * CI workflow build-deps: libavcodec-dev / libavformat-dev /
+    libavutil-dev (Debian stock 7.1.3) → ffmpeg-v4l2-request-fourier
+    (provides /opt/fourier/include + .pc files).
+  * Wire protocol unchanged.  No daedalus-v4l2-dkms bump.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 21:30:00 +0000
+
+daedalus-v4l2 (0.1.0+r39+g3bc0da1-1) bookworm trixie; urgency=medium
+
+  * Bump to 3bc0da1 — picks up daedalus-v4l2 PR #15.  Per-frame
+    `decoder: OK ...` log line gains `decode_us=N` (libavcodec
+    send_packet + receive_frame wall-clock cost in microseconds).
+    New `decoder stats` summary line every 60 decoded frames with
+    codec, fps, avg decode_us, MBs/s throughput, B/MB bitrate.
+  * Pure observability — no decode-path behaviour change.
+    Establishes baseline metrics for the substitution work in
+    daedalus-v4l2#11 step 2 (replacing libavcodec primitives with
+    daedalus-fourier kernels one cycle at a time).
+  * On Pi CM5 / bbb 720p H.264 baseline: ~4 ms decode_us / 24 fps
+    / 90 K MBs/s — workload is well under 1 % of any single
+    daedalus-fourier kernel's NEON ceiling.
+  * Wire protocol unchanged.  No daedalus-v4l2-dkms bump needed.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 18:30:00 +0000
+
+daedalus-v4l2 (0.1.0+r37+g77e14e5-1) bookworm trixie; urgency=medium
+
+  * Bump to 77e14e5 — picks up daedalus-v4l2 PRs #12 + #13.
+  * #12 (LOW_DELAY half-measure): the daemon now sets
+    AV_CODEC_FLAG_LOW_DELAY on the H.264 AVCodecContext so libavcodec
+    emits frames in decode order ~99% of the time (a few stragglers
+    at GOP boundaries when the stream's SPS num_reorder_frames
+    overrides the flag).  Visible improvement vs the 2-1-4-3
+    pair-swap on Firefox YouTube + mpv playback; not a permanent
+    fix (see #11 for the architectural plan).
+  * #13 (daedalus-fourier linkage): the daemon now pkg-config-links
+    against the daedalus-fourier kernel library (marfrit/
+    daedalus-fourier) and logs substrate availability at startup.
+    No kernels dispatched yet — this is the build-time / link-time
+    foundation for the H.264 daemon-rewrite plan in #11
+    (substituting daedalus-fourier IDCT 4×4 / IDCT 8×8 / luma
+    deblock primitives for libavcodec's per-MB pixel math, one
+    cycle at a time, measuring CPU saved per substitution).
+  * Build-deb.sh now fetches + builds + installs daedalus-fourier
+    (pinned at d87239d, marfrit/daedalus-fourier PR #1) into a
+    per-build temp prefix, then builds the daemon with
+    PKG_CONFIG_PATH pointing at it.  daedalus-fourier is
+    statically linked into the daemon binary, so the resulting
+    .deb has no new runtime deps.  Requires libvulkan-dev +
+    glslang-tools on the CI runner (the daedalus-fourier benches
+    already needed those).
+  * Wire protocol unchanged — DAEDALUS_PROTO_VERSION stays at 0.
+    No daedalus-v4l2-dkms bump needed.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 16:30:00 +0000
+
+daedalus-v4l2 (0.1.0+r33+g5d8b436-1) bookworm trixie; urgency=medium
+
+  * Bump to 5d8b436 — reverts daedalus-v4l2 PRs #7 + #8 (the parking
+    design that broke libva-v4l2-request-fourier's 1:1 CAPTURE
+    contract; see daedalus-v4l2#9 + #10).  After daemon-r28+g79256dc
+    landed, mpv (--hwdec=vaapi-copy) failed pre-playing with
+    "Unable to dequeue buffer: Resource temporarily unavailable" /
+    "Failed to end picture decode" because the daemon parked CAPTURE
+    buffers waiting for libavcodec to release H.264 B-frames in
+    display order — violating the V4L2 stateless 1:1 contract.
+    Firefox tolerated the mess (visible "2 1 4 3" pair-swap); mpv
+    bailed.
+  * This bump restores f0d4186-equivalent behaviour, plus PR #4
+    (cosmetic H.264 DECODE_MODE / START_CODE menu controls).  PR #7
+    + PR #8 wire-protocol additions (src_pts / output_src_pts /
+    RESP_FRAME flags) are reverted — DAEDALUS_PROTO_VERSION drops
+    back from 1 → 0.  Lock-step install with daedalus-v4l2-dkms
+    0.1.0+r33+g5d8b436 REQUIRED.
+  * Visible regression: H.264 B-frame streams in Firefox revert to
+    the original "2 1 4 3 6 5" pair-swap visual.  The proper fix
+    (concurrent in-flight requests in daemon + display-order reorder
+    in libva-v4l2-request-fourier) is tracked at daedalus-v4l2#11.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 14:50:00 +0000
+
 daedalus-v4l2 (0.1.0+r28+g79256dc-1) bookworm trixie; urgency=medium
 
   * Bump to 79256dc — H.264 B-frame display reorder fix (closes

debian/ffmpeg-v4l2-request-fourier/0003-h264-idct4-daedalus-fourier.patch

@@ -0,0 +1,137 @@
+From f760c0541586f43334c02611fcb4c212c08ad576 Mon Sep 17 00:00:00 2001
+From: Markus Fritsche <mfritsche@reauktion.de>
+Date: Thu, 21 May 2026 21:40:22 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 4x4 IDCT through
+ daedalus-fourier
+MIME-Version: 1.0
+Content-Type: text/plain; charset=UTF-8
+Content-Transfer-Encoding: 8bit
+
+H264DSPContext.idct_add (called per 4x4 block from the intra-4x4
+decode path in h264_mb.c) now dispatches through
+daedalus_recipe_dispatch_h264_idct4 instead of ff_h264_idct_add_neon.
+
+The recipe layer picks the substrate; for cycle 6 (H.264 IDCT 4x4)
+the recipe is CPU NEON, so this is effectively a NEON-to-NEON
+substitution with one extra dispatch call and recipe-table lookup.
+Provides the first end-to-end exercise of the daedalus-fourier
+kernel pack inside the libavcodec.so decode hot path; follow-up
+patches wire IDCT 8x8, luma-v deblock, and qpel mc20.
+
+The library context is process-global, lazily initialised under
+pthread_once on first call.  We pick the no-QPU constructor because
+libavcodec.so is loaded into arbitrary host processes
+(firefox-fourier, mpv-fourier, daedalus_v4l2_daemon, ...) and we
+cannot assume the host has a usable Vulkan instance.  Higher cycles
+(deblock luma-v, MC) that benefit from the QPU will provision their
+own recipe-selected context once that path is wired.
+
+Bulk paths (idct_add16, idct_add16intra, idct_add8 — used for
+non-intra4x4 macroblocks) remain on the stock NEON .S implementations
+and will be batched through daedalus_recipe_dispatch_h264_idct4 with
+n_blocks>1 in a follow-up.
+
+Bit-exact against ff_h264_idct_add_neon (daedalus-fourier cycle 6
+green; see marfrit/daedalus-fourier/CYCLE_LOGS.md).
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2.
+---
+ libavcodec/aarch64/Makefile               |  3 +-
+ libavcodec/aarch64/h264_idct_daedalus.c   | 49 +++++++++++++++++++++++
+ libavcodec/aarch64/h264dsp_init_aarch64.c |  3 +-
+ 3 files changed, 53 insertions(+), 2 deletions(-)
+ create mode 100644 libavcodec/aarch64/h264_idct_daedalus.c
+
+diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
+index 41ab025..7b95fb1 100644
+--- a/libavcodec/aarch64/Makefile
++++ b/libavcodec/aarch64/Makefile
+@@ -3,7 +3,8 @@ OBJS-$(CONFIG_AC3DSP)                   += aarch64/ac3dsp_init_aarch64.o
+ OBJS-$(CONFIG_FDCTDSP)                  += aarch64/fdctdsp_init_aarch64.o
+ OBJS-$(CONFIG_FMTCONVERT)               += aarch64/fmtconvert_init.o
+ OBJS-$(CONFIG_H264CHROMA)               += aarch64/h264chroma_init_aarch64.o
+-OBJS-$(CONFIG_H264DSP)                  += aarch64/h264dsp_init_aarch64.o
++OBJS-$(CONFIG_H264DSP)                  += aarch64/h264dsp_init_aarch64.o \
++                                           aarch64/h264_idct_daedalus.o
+ OBJS-$(CONFIG_HUFFYUVDSP)               += aarch64/huffyuvdsp_init_aarch64.o
+ OBJS-$(CONFIG_H264PRED)                 += aarch64/h264pred_init.o
+ OBJS-$(CONFIG_H264QPEL)                 += aarch64/h264qpel_init_aarch64.o
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+new file mode 100644
+index 0000000..538d223
+--- /dev/null
++++ b/libavcodec/aarch64/h264_idct_daedalus.c
+@@ -0,0 +1,49 @@
++/*
++ * H.264 4x4 IDCT + add — daedalus-fourier substitution shim.
++ *
++ * Routes H264DSPContext.idct_add through
++ * daedalus_recipe_dispatch_h264_idct4 instead of ff_h264_idct_add_neon.
++ * The recipe layer picks the substrate (CPU NEON by default for
++ * cycle 6; future cycles may dispatch to V3D opportunistically).
++ *
++ * FFmpeg's 4x4 block memory layout matches daedalus's column-major
++ * convention: block[r + 4*c] = coefficient at (row r, col c).  Both
++ * sides destructively zero the block after the transform.
++ *
++ * The library context is process-global and lazily initialised under
++ * pthread_once.  We pick the no-QPU constructor here because
++ * libavcodec.so is loaded into arbitrary host processes
++ * (firefox-fourier, mpv-fourier, daedalus_v4l2_daemon, ...) and we
++ * cannot assume the host has a usable Vulkan instance.  Higher cycles
++ * (deblock, MC) that benefit from the QPU initialise their own
++ * recipe-selected context once that path is wired.
++ */
++
++#include <pthread.h>
++#include <stddef.h>
++#include <stdint.h>
++
++#include <daedalus.h>
++
++#include "libavutil/attributes.h"
++#include "libavcodec/h264dsp.h"
++
++static daedalus_ctx     *g_dctx;
++static pthread_once_t    g_dctx_once = PTHREAD_ONCE_INIT;
++
++static void daedalus_ctx_init_once(void)
++{
++    g_dctx = daedalus_ctx_create_no_qpu();
++}
++
++void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride);
++
++void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
++{
++    static const daedalus_h264_block_meta meta = { .dst_off = 0 };
++
++    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
++
++    daedalus_recipe_dispatch_h264_idct4(g_dctx, dst, (size_t)stride,
++                                        block, 1, &meta);
++}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+index c684574..b993df2 100644
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c
++++ b/libavcodec/aarch64/h264dsp_init_aarch64.c
+@@ -66,6 +66,7 @@ void ff_biweight_h264_pixels_4_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride
+                                     int weights, int offset);
+ 
+ void ff_h264_idct_add_neon(uint8_t *dst, int16_t *block, int stride);
++void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride);
+ void ff_h264_idct_dc_add_neon(uint8_t *dst, int16_t *block, int stride);
+ void ff_h264_idct_add16_neon(uint8_t *dst, const int *block_offset,
+                              int16_t *block, int stride,
+@@ -139,7 +140,7 @@ av_cold void ff_h264dsp_init_aarch64(H264DSPContext *c, const int bit_depth,
+         c->biweight_pixels_tab[1] = ff_biweight_h264_pixels_8_neon;
+         c->biweight_pixels_tab[2] = ff_biweight_h264_pixels_4_neon;
+ 
+-        c->idct_add        = ff_h264_idct_add_neon;
++        c->idct_add        = ff_h264_idct_add_daedalus;
+         c->idct_dc_add     = ff_h264_idct_dc_add_neon;
+         c->idct_add16      = ff_h264_idct_add16_neon;
+         c->idct_add16intra = ff_h264_idct_add16intra_neon;
+-- 
+2.47.3
+

debian/ffmpeg-v4l2-request-fourier/build-deb.sh

@@ -33,7 +33,15 @@ FFMPEG_VERSION=8.1
 # epoch 2 matches Debian's stock ffmpeg (currently 7:7.1.x in trixie);
 # +rfourier suffix to avoid colliding with upstream/Debian rebuilds.
 PKGVER=2:${FFMPEG_VERSION}+rfourier+gb57fbbe
-PKGREL=2  # pkgrel=2 — Path A move to /opt/fourier prefix (2026-05-19)
+PKGREL=5  # pkgrel=5 — H.264 IDCT 4x4 daedalus-fourier substitution; skip past
+          # an orphan -4 .deb sitting in the apt pool that made
+          # check-already-published.sh's `pool_ver ge source_full` short-
+          # circuit the previous -3 build (PR #76).  (2026-05-21)
+
+# daedalus-fourier pin — first kernel substitution in libavcodec (cycle 6
+# H.264 IDCT 4x4).  Same SHA as the daedalus-v4l2 daemon already ships
+# inline; rev in lockstep with the daemon when the public API rolls.
+DAEDALUS_FOURIER_COMMIT=d87239d8172307d9a1b93c95cbed116d175b85cc
 
 HERE=$(dirname "$(readlink -f "$0")")
 
@@ -57,6 +65,34 @@ fi
 # Apply patches (same as Arch).
 patch -Np1 -i "$HERE/0001-libudev-bypass-fallback.patch"
 patch -Np1 -i "$HERE/0002-nv15-to-p010-unpack.patch"
+patch -Np1 -i "$HERE/0003-h264-idct4-daedalus-fourier.patch"
+
+# --- daedalus-fourier: fetch + build static .a with PIC, install to a
+# per-build prefix; libavcodec.so links it into the shared object so
+# H264DSPContext.idct_add (and follow-up kernels) dispatch through the
+# daedalus recipe layer instead of the in-tree NEON .S code. ---
+#
+# PIC is mandatory — the static .a is linked into a .so, so all object
+# code must be relocatable.  Vulkan is PUBLIC-linked by daedalus_core
+# (queryable QPU substrate); we add libvulkan1 to Debian Depends below
+# so dlopen of libavcodec.so.62 succeeds on stock trixie.
+FOURIER_PREFIX=$work/fourier-prefix
+mkdir -p "$FOURIER_PREFIX"
+
+pushd "$work" >/dev/null
+curl --connect-timeout 10 --max-time 600 --retry 3 --retry-delay 5 -sSLfo daedalus-fourier.tar.gz \
+    "https://git.reauktion.de/marfrit/daedalus-fourier/archive/${DAEDALUS_FOURIER_COMMIT}.tar.gz"
+tar xzf daedalus-fourier.tar.gz
+pushd daedalus-fourier >/dev/null
+cmake -B build -G Ninja \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+    -DCMAKE_INSTALL_PREFIX="$FOURIER_PREFIX"
+cmake --build build --target daedalus_core
+cmake --install build
+popd >/dev/null
+popd >/dev/null
+cd "$work/FFmpeg"
 
 # Configure with Arch-parity flags.  Drops the same set of features
 # (X11, AMF, CUDA, FireWire, AviSynth, Bluray, OpenMPT, JPEG-XL,
@@ -73,6 +109,9 @@ patch -Np1 -i "$HERE/0002-nv15-to-p010-unpack.patch"
     --mandir=/opt/fourier/share/man \
     --extra-ldexeflags='-Wl,-rpath,/opt/fourier/lib' \
     --extra-ldsoflags='-Wl,-rpath,/opt/fourier/lib' \
+    --extra-cflags="-I${FOURIER_PREFIX}/include" \
+    --extra-ldflags="-L${FOURIER_PREFIX}/lib" \
+    --extra-libs="-ldaedalus_core -lvulkan -lpthread" \
     --disable-debug \
     --disable-static \
     --disable-doc \
@@ -147,6 +186,7 @@ Priority: optional
 Architecture: arm64
 Depends: libc6,
          libdrm2,
+         libvulkan1,
          libfontconfig1,
          libfreetype6,
          libfribidi0,

debian/ffmpeg-v4l2-request-fourier/debian/changelog

@@ -1,3 +1,48 @@
+ffmpeg-v4l2-request-fourier (2:8.1+rfourier+gb57fbbe-5) bookworm trixie; urgency=medium
+
+  * pkgrel-only bump (3 → 5) to force a rebuild of the H.264 IDCT 4x4
+    daedalus-fourier substitution that landed in marfrit-packages PR
+    #76.  An orphan -4 .deb already sat in the apt pool (dated
+    2026-05-19, no matching source commit in main); CI's
+    check-already-published.sh compares with `dpkg --compare-versions
+    pool_ver ge source_full`, which short-circuited PR #76's -3
+    build.  Skipping past -4 lets the CI workflow actually publish the
+    substitution.
+  * No source code change beyond PKGREL and this changelog entry.
+    Substitution + control + build-deb.sh wiring stay as PR #76 left
+    them.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 21:30:00 +0000
+
+ffmpeg-v4l2-request-fourier (2:8.1+rfourier+gb57fbbe-3) bookworm trixie; urgency=medium
+
+  * Add 0003-h264-idct4-daedalus-fourier.patch — H264DSPContext.idct_add
+    (per-block 4x4 IDCT, called from the intra-4x4 decode path in
+    libavcodec/h264_mb.c) now dispatches through
+    daedalus_recipe_dispatch_h264_idct4 instead of
+    ff_h264_idct_add_neon.  First end-to-end exercise of the
+    daedalus-fourier kernel pack inside libavcodec.so on the
+    production decode hot path (daedalus-v4l2#11 step 2 — cycle 6
+    H.264 IDCT 4x4, NEON-by-recipe).
+  * build-deb.sh: fetches + builds daedalus-fourier (pinned at
+    d87239d, lockstep with the daemon's static link) with
+    -fPIC into a per-build temp prefix, then passes
+    --extra-cflags=-I.../include --extra-ldflags=-L.../lib
+    --extra-libs="-ldaedalus_core -lvulkan -lpthread" to FFmpeg
+    configure.  Static-linked into libavcodec.so.62.
+  * Bulk paths (idct_add16 / idct_add16intra / idct_add8) remain on
+    the stock NEON .S code and will be batched through
+    daedalus_recipe_dispatch_h264_idct4 with n_blocks>1 in a
+    follow-up.  Cycles 7/8/9 (IDCT 8x8 / luma-v deblock / qpel mc20)
+    land in subsequent patches.
+  * Depends gains libvulkan1 — daedalus_core PUBLIC-links Vulkan
+    (queryable QPU substrate); the no-QPU constructor still works,
+    but the loader refuses libavcodec.so.62 at dlopen time without
+    libvulkan.so.1 present.
+  * No ABI change; SONAMEs stay 62/62/60.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 20:00:00 +0000
+
 ffmpeg-v4l2-request-fourier (2:8.1+rfourier+gb57fbbe-1) bookworm trixie; urgency=medium
 
   * Initial Debian packaging for the Kwiboo FFmpeg fork with V4L2